Proceedings of SALT 25: 184–203, 2015
De re tenses and trace conversion∗
Moshe E. Bar-Lev
The Hebrew University of Jerusalem
Abstract I suggest a quantificational account for tenses in which the seemingly
peculiar behavior of tenses that are interpreted de re (most notably the double access
reading of English present-under-past sentences) falls out from a general Trace
Conversion rule that applies to moved quantifiers, as in Fox 2002. I propose that de
re tenses involve movement (following Ogihara 1989), and that the first argument
of tenses is a property of times which characterizes the set of times that include
the local evaluation time, such that the application of Trace Conversion to moved
tenses yields an inclusion requirement with respect to the local evaluation time of the
base position. Unlike previous analyses (Ogihara 1989; Abusch 1997), the current
analysis predicts that a de re interpretation of a tense (past or present) involves
inclusion of the attitude time. This is supported by the availability of simultaneous
readings for past-under-past sentences in non-SOT languages such as Hebrew, and
the unavailability of ‘mixed’ (simultaneous and backward-shifted) readings for
past-under-past constructions under universal quantification.
Keywords: semantics of tense, quantification, trace conversion, de re, sequence of tense,
double access, Hebrew
1
Introduction
Cases where embedded tenses seem to be interpreted de re, namely indexically, are
puzzling from the point of view of the semantics of tenses. The problem has been
mostly noted for the double access reading of English present-under-past sentences
like (1): Whenever a tense is interpreted as indexical, it seems that it also contributes
a relative meaning, as if it were both an indexical and a relative tense at the same
time. Intuitively, the embedded tense in (1) says something both about the speaker’s
∗ This paper is based on my MA thesis which was supervised by Nora Boneh, to whom I am grateful
for very helpful discussions and comments on various stages of this work. I have benefited a lot
from discussions with Daniel Altshuler, Elitzur Bar-Asher Siegal, Luka Crnič, Danny Fox, Yosef
Grodzinsky, Yael Sharvit, Itai Bassi, David Kashtan and Daniel Margulis. Thanks to all the faculty
and students at the Language, Logic and Cognition Center, to the participants at the SOT meetings
held at the LLCC and to the audiences and abstract reviewers of SALT 25, GLOW 38, the linguistics
colloquium at Ben-Gurion University and the linguistics colloquium at Tel Aviv University. All
mistakes are my own. The research was partly funded by ISF grant 1366/14 to Nora Boneh.
©2015 Bar-Lev
De re tenses and trace conversion
“now”, the utterance time, hence indexical meaning; and about John’s “now”, the
attitude time, hence relative meaning. However, there has been disagreement about
the nature of that relative meaning.
(1)
John thought that Mary is pregnant.
This paper aims to argue that under close scrutiny it can be seen that the relative
component that de re tenses contribute is a kind of present. That is, de re tenses
contribute a relative meaning of inclusion of the attitude time. A further claim that
will be made here is that this puzzling behavior teaches us something about the semantics of tenses: that they are quantificational, and that they quantify over the set of
times that include the local evaluation time. Together with an independently needed
Trace Conversion rule for the syntax-semantics interface (following Fox 2002), this
enables us to predict that whenever a tense moves, its lower copy is interpreted as a
definite description which presupposes inclusion of its local evaluation time. And
since de re readings of tenses will be analyzed (following Ogihara 1989) as involving
movement, the relative meaning of inclusion for de re tenses will be predicted.
The paper is structured as follows: In Section 2 I present the problem with the
analysis of double access, and the idea that it is an instance of de re interpretation of
tenses. In Section 3 I present my analysis which is based on Trace Conversion. In
Section 4 I argue that the current proposal fares better than previous de re analyses
of tenses, in light of de re interpretations of past-under-past sentences. I show that
the proposed analysis predicts that every de re tense contributes a relative meaning
of inclusion of the attitude time (Section 4.1), and support this prediction by the
availability of simultaneous readings for past-under-past sentences in non-SOT
languages such as Modern Hebrew (Section 4.2), and the unavailability of ‘mixed’
(simultaneous and backward-shifted) readings for past-under-past structures under
universal quantification (Section 4.3) . Section 5 summarizes and discusses some
problems and an alternative analysis.
2
2.1
Double access
Data
A present complement clause embedded under a past attitude verb in English, as in
(1), has only the double access reading (DAR). DAR requires the embedded event
not only to hold at the attitude time (AT = the local evaluation time of the embedded
clause, John’s “now” in (1)) but to encompass the utterance time (UT = the local
evaluation time of the matrix clause, the speaker’s “now” in (1)) too.1
1 This is not an exact characterization of DAR, but it suffices for the discussion. For detailed discussions
about the nature of DAR see Ogihara 1989; Abusch 1997; Bary & Altshuler 2015; Klecha 2014.
185
Bar-Lev
Starting out from the assumption that tenses are always relative — namely that
they are evaluated with respect to the local evaluation time of the clause in which
they appear — this is surprising: Why would the event in the embedded clause ‘care’
about the utterance time at all? This is illustrated in (2).
(2)
Relative tenses:
John thought
?
Utterance time (UT)
the speaker’s “now”
that Mary is pregnant
Attitude time (AT)
John’s “now”
In order to see that the utterance time plays a crucial role in the interpretation of
sentences such as (1), consider the following well-observed contrast:
(3)
Double access:
a. Yesterday, John thought that Mary is pregnant.
b. #Two years ago, John thought that Mary is pregnant.
Presumably, (3b) is infelicitous since it posits a requirement on the time of which
John believed Mary’s pregnancy to hold that clashes with our world knowledge in
which human pregnancies cannot last two years.
If so, assuming that the embedded present in (1) is simply relative we are not
able to explain how the utterance time becomes relevant to the running time of the
embedded event. On the other hand, assuming that it is indexical, namely evaluated
with respect to the utterance time, won’t be enough either. We would just replace
the question of how the utterance time is relevant for the embedded event with the
question of how the attitude time is relevant for it.
(4)
Indexical tenses:
John thought
Utterance time (UT)
the speaker’s “now”
that Mary is pregnant
?
Attitude time (AT)
John’s “now”
That the attitude time cannot be ignored in the interpretation of (1) can be seen in
that (1) has no reading along the lines of (5).
(5)
John thought that Mary would be pregnant now.
On the face of it, there are two ways to treat the embedded present in (1), as indexical
or as relative, but neither is adequate on its own. That is, in order to account for the
DAR of present-under-past sentences like (1), two things should be derived: (i) an
indexical component for DAR, namely sensitivity of the embedded event to UT, and
(ii) a relative component for DAR, namely sensitivity of the embedded event to AT.
186
De re tenses and trace conversion
2.2 De re approaches
A common approach to facing this problem, mainly due to Ogihara (1989) and
Abusch (1997), is that the embedded present in (1) is interpreted de re.2 Put simply,
it is an indexical tense. However, as we have seen this only gives us the means to
provide an indexical component for DAR. But what about a relative component?
Two main answers were given to this question within de re analyses of DAR:
one by Abusch (1997), and the other by Ogihara (1989). For now I will only briefly
describe their answers, and in Section 4 I will discuss them in somewhat more detail.
Abusch’s answer is that there is a general restriction on the interpretation of
tenses which she calls the Upper Limit Constraint (ULC). Following Heim’s (1994)
implementation, it prohibits assigning to a tense node a time that is later than its
local evaluation time.
Assuming that the embedded present in (1) is a de re tense, the ULC entails
effectively that (1) can only be felicitous if John thought of Mary’s pregnancy time
as being before or including his “now”, i.e., if it is not interpreted as being after AT.
However, the ULC-based approach faces some conceptual as well as empirical
problems. One empirical problem will be discussed in Section 4.3. For now I
will only point out the conceptual problems it raises. First, the ULC is an ad hoc
constraint. Even though it provides a relative component for de re tenses, it is not
clear what other motivation this kind of formal constraint might have. Second, the
ULC forces us to treat future as something different than present and past — i.e., not
as a tense — since future gives us a time that is later than the local evaluation time,
which is what the ULC prevents. Even though saying that future is not a tense might
not be such a bad idea, the ULC leaves us no other option and as such it limits our
theoretical possibilities.
Escaping these kinds of problems and not utilizing the dubious ULC, Ogihara’s
(1989) Copy-based approach suggests the DAR is the result of tense movement that
leaves a copy, which is in line with the Copy theory of movement.
Following Ogihara (1989), I will suggest that de re tenses are the result of
movement that leaves a copy in the embedded clause. The higher copy of the tense
provides an indexical component, and the lower copy is responsible for a relative
component for de re tenses.
The claim that I put forward in this paper is that the lower copy of de re tenses
provides a relative component which requires the embedded event to include the
attitude time. This generalizes to any de re interpretation of tenses, even de re past
tense. The generalization is in (6).
2 I set aside the question of what makes the de re interpretation of the embedded present in (1)
obligatory, which seems to be a mystery. See Abusch 1997 and Klecha 2014.
187
Bar-Lev
(6)
I NCLUSION G ENERALIZATION :
A de re tense is interpreted as including the local evaluation time of its base
position (AT).
Section 3 presents an analysis that predicts (6). Section 4 adduces evidence in favor
of (6).
3
Proposal: A Trace-Conversion-based approach
We have seen that de re approaches suggest that the embedded present in presentunder-past sentences is interpreted de re, and that Ogihara suggested that this interpretation comes about by movement, such that the higher copy provides an indexical
component and the lower copy is responsible for a relative component. The goal
of this section is to motivate and provide an analysis in which the interpretation of
the lower copy of de re tenses is always the same: They are interpreted as a definite
description that presupposes inclusion of the attitude time, thus conforming to (6).
3.1
Trace Conversion for quantificational tenses
As a first step towards an analysis, I want to present an interpretability problem that
arises if we assume that tenses are quantificational, that they may be interpreted de re,
and that de re tenses involve copy movement as in Ogihara’s Copy-based approach.
Then I present my analysis that solves that problem using Trace Conversion and the
assumption that tenses quantify over the set of times that include the local evaluation
time. Later I show that the proposed analysis also predicts (6).
There is a long standing debate whether tenses are to be analyzed as quantificational or as pronominal, namely as referring expressions. Partee (1973) has claimed
that tenses should be treated as pronominal since they behave as if they refer to times.
However, Stalnaker (as cited by von Fintel & Heim 2011), and later Ogihara (1989)
proposed that tenses are still quantificational and that similarly to other quantifiers
they quantify over a contextually provided domain of times.3
Let us assume a quantificational semantics for tenses. The question that this
section discusses is this: Given this assumption, and the assumption that tenses can
move and leave a copy behind as in Ogihara’s Copy-based approach, then what is the
interpretation of a tense, i.e., a temporal quantifier, that has copies in two different
places in a single LF?
Specifically, take the LF of (1) in which the embedded present has moved to get
a de re interpretation. In the Copy-based approach, a copy is left behind as in (7).
3 In the following discussion I neglect the contextual restriction, which does not play a crucial role
here.
188
De re tenses and trace conversion
What is the interpretation of such an LF?
(7)
PRES3
λ 7 PAST1 John think PRES7 Mary be-pregnant
This relates to a general problem with the interpretation of quantifiers that move
within the Copy theory of movement: If we want to account for the reading of (8)
where every girl takes higher scope than a boy we would want to move every girl
up as in (8a). But leaving a full copy of a quantifier phrase of type het,ti in two
places in the structure yields an uninterpretable structure. The Trace Conversion rule
suggested by Fox (2002), in (9), is designed to solve such problems. It converts the
lower copy into a definite description of type e, as in (8b).
(8)
A boy kissed every girl.
a. ?every girl λ 1 a boy kissed [every girl]1
b. every girl λ 1 a boy kissed [the girl 1]
(9)
T RACE C ONVERSION:
a. Variable Insertion (VI): (Det) Pred → (Det) [Pred λ y(y = x)]
b. Determiner Replacement (DR):
(Det) [Pred λ y(y = x)] → the [Pred λ y(y = x)]
(Fox 2002: 67)
The exact same problem with the interpretation of (8a) would apply to structures
such as in (7), if tenses are quantifiers, that is, if the type of PRES7 is hit,ti.4 A
natural move is to apply Trace Conversion to (7) too. Unfortunately, it turns out that
the Trace Conversion rule cannot be applied to quantificational tenses as usually
defined. Consider a simple quantificational tense semantics as in (10).
(10)
a.
b.
JPRESK(ti )(Phi,ti ) = 1 iff ∃t 0 [t 0 ⊇ t ∧ P(t 0 ) = 1]
JPASTK(ti )(Phi,ti ) = 1 iff ∃t 0 [t 0 < t ∧ P(t 0 ) = 1]
An immediate problem arises: The first argument of tenses in (10) is a time and not a
property of times, unlike nominal quantifiers. Applying the Trace Conversion rule in
such a case wouldn’t do any good: The converted quantifier would end up meaning
something like “the today”, since the argument of the would be a time rather than a
property of times.
The question can be asked as follows: If tenses are quantifiers, just as nominal
quantifiers, why do they take a time and not a property of times? What is the reason
4 In Ogihara 1989 this problem is avoided by assuming that quantification is not part of the core
semantics of tenses, but rather tenses are properties of times/states, and existential closure is done
higher up. Additionally, his proposal is based on a Davidsonian approach with reference to states,
which he claims provides the right results for capturing the tricky DAR associated with present-underpast sentences. To keep the presentation traceable I only use his Copy movement idea, similarly to
what has been done in Ogihara & Sharvit 2012.
189
Bar-Lev
that nominal quantifiers are of type het, het,tii whereas temporal quantifiers are of
type hi, hit,tii rather than hit, hit,tii?
Therefore, I submit that tenses are indeed of type hit, hit,tii, similarly to nominal
quantifiers. What should be the property they quantify over? There are a few
possibilities here. I suggest that this is the property of including the local evaluation
time, since that property would let us predict (6).
I use a predicate T , a function of type hi, iti, which is defined as follows.
(11)
JT K = λt.λt 0 .t 0 ⊇ t
T takes the local evaluation time as its argument. Given a local evaluation time t,
JT K(t) is the characteristic function of the set of all times that include t.
(12)
JT K(t) = λt 0 .t 0 ⊇ t
The kind of function given by JT K(t) is a suitable candidate for a predicate to be
taken by the temporal quantifier. What the quantifiers take is a predicate of type
hi,ti: a property of times, which is the property of including the evaluation time.
Before we discuss the way in which the semantics of tenses should be amended
in order to work with this kind of argument, note that the application of Trace
Conversion to tenses is now possible. Since all tenses take the property of including
the evaluation time as their argument — that is, an argument which very much
resembles a present tense — the application of Trace Conversion would strip them
of their inherent meanings and leave intact only the ‘presentness’ of their argument.
Assume that we have a structure with a present tense, the first argument of which
is [T t ∗ ], namely the set of times that include the utterance time, as in (13a). After
QRing it as in (13b), we can apply Trace Conversion in (13c) and (13d), to derive
the meaning of the lower copy, which is now of type i, as in (13e).
(13)
a.
b.
c.
d.
e.
. . . [φ PRES [T t ∗ ]] . . .
[PRES [T t ∗ ]] λ x . . . [φ PRES [T t ∗ ]] . . .
[PRES [T t ∗ ]] λ x . . . [φ PRES [[T t ∗ ] λ y(y = x)]] . . .
[PRES [T t ∗ ]] λ x . . . [φ the [[T t ∗ ] λ y(y = x)]] . . .
Jφ K = the t s.t. t ⊇ t ∗ and t = x
QR
VI
DR
The meaning we get for the lower Copy after Trace Conversion in (13e) is independent of whether it was a past or a present to begin with: Since Trace Conversion
replaces the tense with the, the same result of inclusion would be achieved if it were
a past tense. This seemingly strange result is what lets us predict (6), and it will
become crucial later, when we discuss past-under-past sentences (Section 4).
I suggest the semantics of tenses in (14).
190
De re tenses and trace conversion
(14)
a.
b.
JPRESK(Thi,ti )(Phi,ti ) = 1 iff ∃t[T (t) ∧ P(t) = 1]
JPASTK(Thi,ti )(Phi,ti ) = 1 iff ∃t[∃t 0 [T (t 0 ) ∧ t < t 0 ] ∧ P(t) = 1]
I assume that tenses are relative: T takes the local evaluation time as its argument,
and the result of applying T to the local evaluation time, namely the set of times
that include the local evaluation time, is what tenses take as their first argument (T
in (14)).5
Given the set of times that include the local evaluation time, the definition of
PRES in (14a) is simple: It yields truth in case there is a time which is within this set
and the given proposition is true at that time. The definition of PAST in (14b), given
such a set, is more complex: It yields truth in case there is a time (t) which is before
some time (t 0 ) within this set, and the proposition is true at that time (t).6
Before deriving the puzzling DAR of (1), let us see how this works in simple
cases. I assume a system in which world and time variables are represented in the
syntax, and predicates and clauses are of type hs, iti (cf. Percus 2000; Kusumoto
2005). Truth for utterances is defined as in (15), following von Fintel & Heim’s
(2011: 104) definition, modified to incorporate times:7
(15)
An utterance of a sentence φ in world w at time t is true iff Jφ K(w)(t) = 1.
A tenseless clause such as “Mary be-pregnant” has the representation in (16).
(16)
λ w1 λt1 Mary be-pregnantw1 ,t1
Not having tenses, the world and time arguments taken by the predicate be-pregnant
5 For any evaluation time ti there is a unique function Ti such that Ti = JT K(ti ), which is the argument
of tenses whenever they appear in a position where ti is the local evaluation time. Therefore in simple
cases (i.e., when no movement is involved) the definitions of tenses in (14) give equivalent results to
the standard definitions in (10), since the equivalences in (i) hold:
(i)
Given a local evaluation time ti and the function Ti s.t. Ti = JT K(ti ):
a.
∃t[Ti (t) ∧ P(t)] ⇔ ∃t[t ⊇ ti ∧ P(t)]
b.
∃t[∃t 0 [Ti (t 0 ) ∧ t < t 0 ] ∧ P(t)] ⇔ ∃t 0 [t 0 < ti ∧ P(t 0 )]
6 This definition of PAST renders it a non-conservative quantifier. See Section 5.
7 I choose such a system mainly for convenience. The desired results can be derived otherwise, for
example utilizing a distinguished variable 0 for the local evaluation time, as in Heim 1994. This
requires some modifications, see footnote 8.
Additionally, the semantics I assume for think is in (ii).
(ii)
JthinkK(w)(t)(P<s,it> )(x) = 1 iff ∀hw0 ,t 0 i ∈ Doxx (w,t) : P(w0 )(t 0 ) = 1
Where Doxx (w,t) is the set of doxastic alternatives of x in w at t.
191
Bar-Lev
are bound by the local evaluation world and time — w1 and t1 .
A tensed clause as in (17a) has the structure in (17b) and the truth conditions
(given the actual world w∗ and the utterance time t ∗ ) in (18).
(17)
a.
b.
Mary was pregnant.
λ w1 λt1 [PAST [T t1 ]] λt3 Mary be-pregnantw1 ,t3
(18)
J(17b)K(w∗ )(t ∗ ) = 1 iff
∃t[∃t 0 [t 0 ⊇ t ∗ ∧ t < t 0 ] ∧ Mary be-pregnant in w∗ at t] iff
∃t[t < t ∗ ∧ Mary be-pregnant in w∗ at t]
3.2
Application — de re tenses in present-under-past sentences
We can now show how the DAR can be derived for (1). First, the embedded PRES
together with its first argument undergo QR, as in (19a). Since they appear in the
scope of two different local evaluation times, the two copies are not completely
identical: In the lower copy the argument of T is t3 which corresponds to AT,
whereas in the higher copy the argument of T is t1 which corresponds to UT.8
Second, the application of Trace Conversion to the lower copy yields the interpretable structure in (19b). The truth conditions of (19b) given the actual world and
the utterance time are in (19c).
(1)
(19)
John thought that Mary is pregnant.
a.
b.
c.
QR:
λ w1 λt1 [PRES [T t1 ]] λt7 [PAST [T t1 ]] λt2 John thinkw1 ,t2
λ w3 λt3 [PRES [T t3 ]] λt4 Mary be-pregnantw3 ,t4
Trace Conversion:
λ w1 λt1 [PRES [T t1 ]] λt7 [PAST [T t1 ]] λt2 John thinkw1 ,t2
λ w3 λt3 [the [[T t3 ] λ y.y = t7 ]] λt4 Mary be-pregnantw3 ,t4
J(19b)K(w∗ )(t ∗ ) = 1 iff
∃t 00 [t 00 ⊇ t ∗ ∧ ∃t 0 [t 0 < t ∗ ∧ ∀hw,ti ∈ DoxJohn (w∗ )(t 0 ) :
Mary be-pregnant in w at the t 000 s.t. t 000 ⊇ t and t 000 = t 00 ]]
The LF in (19a) corresponds to the desired double access meaning. The relative
8 A potential advantage of using the distinguished variable 0 for the local evaluation time (see footnote
7), is that there would be no need to assume copies which are not identical as in (19), a situation which
has no equivalent in the nominal domain. This requires replacing the first argument of tenses with
T0 , where JT0 K = λt.t ⊇ t0 : Both copies would be [PRES T0 ], and 0 would get different assignments
depending on the clause in which it appears. Such a move requires the further assumption that the
local evaluation time has a special status, i.e., that λ -abstraction over it is not created by QR. This is
what is assumed by von Stechow’s (1995) QR CONVENTION: “The movement index created by QR
is always different from the distinguished index 0”. I thank Uli Sauerland for highlighting this issue.
192
De re tenses and trace conversion
component of DAR is the result of the converted copy downstairs, because of which
the pregnancy time believed by John is presupposed to include AT, i.e., John’s “now”;
and the indexical component of DAR is asserted by the higher copy: The time of
which John believed Mary to be pregnant is extensionally required to include UT.9
Just like in Ogihara’s original Copy-based approach, both the indexical component and the relative component of DAR encode inclusion. In the next section,
which is dedicated to a comparison of the current proposal to previous ones in light
of past-under-past sentences, we will see that the two approaches can be teased apart
when considering such sentences.
4
Substantiation for the analysis: Comparison with previous de re accounts
In this section I consider how the current proposal fares with respect to other de re
accounts. Past-under-past sentences are added to the picture as a way to differentiate
the predictions of the various accounts, and to claim that (6) is the right prediction.
4.1
Setting the scene: Predictions for de re past-under-past sentences
It has been long observed (Jespersen 1924; Ogihara 1989; Abusch 1997; a.o.) that
a past complement clause embedded under a past attitude verb (past-under-past
as in (20)) in English can have either a backward-shifted reading as in (20a) or a
simultaneous reading as in (20b). This is known as Sequence of Tense (SOT).
(20)
John thought that Mary was pregnant.
a. John thought: “Mary was pregnant”.
b. John thought: “Mary is pregnant”.
Backward-shifted reading
Simultaneous reading
I do not deal here directly with explaining the existence of these two readings.
Instead, I focus on the predictions of the various de re analyses for tenses as to the
de re interpretation of past-under-past sentences. First I will go over the prediction
of the current proposal, and then compare it to the predictions of previous analyses.
Recall that the meaning we get for the lower copy in (19) after Trace Conversion
is independent of whether it was PAST or PRES to begin with: Since Trace Conversion
replaces the tense with the, the same result of inclusion for the lower copy would be
achieved if it were a past tense. This is the case in (21) which is the de re LF of (20).
9 An anonymous reviewer claims that as it stands, the truth conditions in (19) are not quite right: Since
t7 appears within the attitude context and bound from outside, it yields a belief of John about a time
that includes the utterance time. However, it is clear that the attitude holder should not have any belief
directly about the utterance time. I hope this can be solved using a concept-generator mechanism as
in Percus & Sauerland 2003 and Charlow & Sharvit 2014 and applying it to the lower copy, such that
the belief is about a time concept applied to t7 rather than about t7 itself.
193
Bar-Lev
(21)
a.
b.
c.
QR:
λ w1 λt1 [PAST [T t1 ]] λt7 [PAST [T t1 ]] λt2 John thinkw1 ,t2
λ w3 λt3 [PAST [T t3 ]] λt4 Mary be-pregnantw3 ,t4
Trace Conversion:
λ w1 λt1 [PAST [T t1 ]] λt7 [PAST [T t1 ]] λt2 John thinkw1 ,t2
λ w3 λt3 [the [[T t3 ] λ y.y = t7 ]] λt4 Mary be-pregnantw3 ,t4
J(21b)K(w∗ )(t ∗ ) = 1 iff
∃t 00 [t 00 < t ∗ ∧ ∃t 0 [t 0 < t ∗ ∧ ∀hw,ti ∈ DoxJohn (w∗ )(t 0 ) :
Mary be-pregnant in w at the t 000 s.t. t 000 ⊇ t and t 000 = t 00 ]]
Crucially, the only place where there is a difference between the truth conditions in
(19c) and (21c) is the doubly-underlined t 00 < t ∗ in (21c) which replaces the t 00 ⊇ t ∗
in (19c). The current proposal thus predicts that the truth conditions of the de re
structures of (20) and (1) would be identical except for the temporal relation with
respect to UT, namely the indexical component. On the other hand, the relative
component in both de re past-under-past and de re present-under-past requires
inclusion. This is exactly what is required for predicting (6), repeated here:
(6)
I NCLUSION G ENERALIZATION :
A de re tense is interpreted as including the local evaluation time of its base
position (AT).
This result comes about given our assumptions: (i) that tenses quantify over the
set of times that include their local evaluation time, (ii) that de re interpretations of
tenses are the result of movement that leaves a copy in the embedded clause, and
(iii) that Trace Conversion applies to that copy, turning it to mean “the time that is
within the set of times that include the local evaluation time (which is identical to
the index created by movement)”.
Let us now compare (6) to the predictions of previous approaches.
Abusch’s ULC-based approach derives a relative component for DAR from the
ULC, which prevents a tense node from being assigned a time that is later than its
local evaluation time. Therefore, the relative component of de re present and that of
de re past would be the same. The ULC-based approach thus predicts (22).
(22)
U PPER L IMIT G ENERALIZATION:
A de re tense is interpreted as preceding or including the local evaluation
time of its base position (AT).
In 4.3 I will argue that (22) is too permissive based on cases where a past-under-past
construction is embedded under universal quantification, and that (6) is preferred.
194
De re tenses and trace conversion
Ogihara’s Copy-based approach suggests that de re readings of tenses are the
result of movement that leaves a copy downstairs, which is in line with the Copy
theory of movement. Since the semantics of the lower copy is what determines the
contribution of the relative component (see footnote 4), the relative component of
de re past would require the embedded event to precede AT. Ogihara’s Copy-based
approach then predicts the following:
(23)
T EMPORAL D IRECTIONALITY G ENERALIZATION:
A de re tense is interpreted as:
a. preceding the local evaluation time of its base position if it’s PAST.
b. including the local evaluation time of its base position if it’s PRES.
In 4.2 I will argue that (23a) does not allow for simultaneous readings of past-underpast in non-SOT languages, whereas given (6) they are naturally explained.
4.2
Simultaneous readings of past-under-past in Hebrew
To test whether de re past contributes a relative component of precedence as in
(23a) or one of inclusion as in (6) we would want to check what happens with
past-under-past sentences: If we could find evidence that de re interpretations of
tenses give rise to simultaneous readings, we could have an argument for (6).
Could we argue that the simultaneous reading in (20b) is an instance of a de re
past? The answer is no. Many analyses10 assume that the embedded past in (20) can
be a zero tense, explaining its simultaneous reading through some kind of deletion
or SOT rules (see Comrie 1986; Stowell 1995, 2007; Kratzer 1998; Schlenker 1999,
2004 and Grønn & von Stechow 2010). Abusch (1997) suggests the simultaneous
reading of were having in (24) as a motivation for such an explanation:
(24)
John decided a week ago that in ten days at breakfast he would say to his
mother that they were having their last meal together. (Abusch 1997: 18)11
The time of having-last-meal in (24) is normally understood to be simultaneous with
the time of having-breakfast and the time of saying, and not to precede any other time
we can think of in this sentence: neither the utterance time nor the decision-time.
This leads to the conclusion that no treatment of this occurrence of past tense as a
real past tense can work, and therefore we need to posit deletion rules for English.
If so, English would be the wrong language to look at when we want to see
if de re past tenses give rise to simultaneous readings, since having deletion rules
we might get such readings in another way. What we should do then is consider
10 Though not all; see e.g., Altshuler & Schwarzschild 2012.
11 The example was originally presented in French in a paper by Kamp and Rohrer.
195
Bar-Lev
languages in which there are no deletion rules, such as Hebrew.
As has been shown by Ogihara & Sharvit (2012), the fact that Hebrew has no
deletion rules is evident in the following Hebrew equivalent to (24), in which the
past on axlu (‘ate’) can only be interpreted as backward-shifted:
(25)
Yoni hexlit lifney šavua lehagid le-ima
šelo beod asara yamim
Yoni decided before week to.tell to-mother his in.more ten days
be-aruxat boker še-hem axlu et ha-aruxa ha-axrona šelahem
in-meal morning that-they ate ACC the-meal the-last their
yaxad.
together
The fact that there is no simultaneous reading available for (26) in the context in (26a)
is expected. Surprisingly, however, when the context is as in (26b), a simultaneous
reading for (26) is quite natural.
(26)
yadati
še-Rina hayta be-herayon!
Knew.1sg that-Rina was in-pregnancy!
‘I knew that Rina was pregnant!’
a. Context: yesterday I saw Rina with a swollen belly.
b. Context: two years ago I saw Rina with a swollen belly.
Sim.: 7
Sim.: 3
But if Hebrew has no deletion rules, what could make any simultaneous reading
possible? Our only candidate is a de re account. The simultaneous reading of (26) in
the context in (26b) could be explained with a de re LF, if we assume (6). But as
Ogihara & Sharvit (2012) pointed out, within the Copy-based approach we cannot
derive simultaneous readings for de re past because of (23a), thus it cannot explain
simultaneous readings of past-under-past in non-SOT languages such as Hebrew.
Moreover, (26) shows that a de re analysis of tenses is needed.12 First, what
determines the availability of a simultaneous reading for (26) is the distance from the
utterance time, which already points at the direction of an interpretation that involves
indexicality, which is what a de re interpretation would yield. Second, consider the
well-observed contrast in (3), repeated here:
(3)
Double access:
a. Yesterday, John thought that Mary is pregnant.
b. #Two years ago, John thought that Mary is pregnant.
12 See Ogihara & Sharvit 2012 for another instance of a simultaneous reading of past-under-past in
Hebrew that involves a mistake of the attitude holder regarding the time he is in, which also points at
the direction of a de re analysis.
196
De re tenses and trace conversion
There is a clear correlation between (26) and (3): A simultaneous reading for
past-under-past in Hebrew requires a setup in which an English present-under-past
wouldn’t be accepted, and vice versa: English present-under-past requires a setup in
which a simultaneous reading for past-under-past in Hebrew wouldn’t be accepted.
This is reminiscent of the correlation between past and present in matrix clauses:
Mary was pregnant isn’t normally accepted when Mary is pregnant is true.
Assuming that both DAR and simultaneous readings of past-under-past in Hebrew are instances of a de re interpretation of a tense, the correlation between (26)
and (3) is expected.13 But if one wants to deny the need for de re tenses and explain
the DAR using an involved semantics for English present which incorporates both
an indexical component and a relative component (as in Gennari 2003; Altshuler &
Schwarzschild 2013 and Bary & Altshuler 2015), one would have to say something
similar about Hebrew past to account for that correlation. However, there is no
evidence that Hebrew past should have an indexical component in its semantics.
4.3
Functional readings of tenses
The prediction of the ULC-based approach in (22) is compatible with simultaneous
readings for de re past, so it can explain the Hebrew data in (26). But (22) predicts
also backward-shifted readings for de re past, against which I argue in this section.
Before I argue against (22), a word about time concepts is in order. Abusch
(1997) utilizes Lewis’ acquaintance relations for tenses, or in Heim’s (1994) terminology: time concepts. The idea is that a de re interpretation of a tense involves
reference to some time, and a time concept by which the attitude holder (John in
our examples) can describe this time to himself. Time concepts can be “yesterday”,
“now”, “tomorrow”, “this year”, or “last time the lights went out”. Given, for example, the time concept “yesterday”, the de re interpretation of (20) is roughly (27)
(where (27a) represents an indexical component and (27b) a relative component):
(27)
The time of which John believed Mary to be pregnant is such that:
a. From the speaker’s perspective it precedes the “now”.
Indexical c.
b. From John’s perspective it’s “yesterday”.
Relative c.
The ULC-based approach then rules out using time concepts such as “tomorrow” or
“a year from now”, because they would lead to a violation of the ULC.14
Consider now the example in (28), in which a past-under-past construction is
13 The exact mechanism by which this correlation should be derived is left to be spelled out. Hopefully,
this would fall out from an analysis of why a PAST-φ sentence has a non-PRES-φ inference. See
Altshuler & Schwarzschild 2012 and references therein.
14 An interesting question arises regarding temporally neutral time concepts such as “on Bill’s 40th
birthday”. Following Heim (1994: fn. 28), Bary & Altshuler (2015) discuss the scenario in (iii):
197
Bar-Lev
embedded under universal quantification over individuals. Just like any other pastunder-past sentence in English, it has two readings: a backward-shifted reading as in
(28a) and a simultaneous reading as in (28b). Importantly, (28a) is a reading of (28)
even with a functional meaning, i.e., if every doctor had another past-time in mind.
However, there is no ‘mixed’ functional reading for (28) in which for some doctors
it was a past-time and for some doctors it was a present-time, as in (28c).15
(28)
Every doctor thought Mary was pregnant.
a. Every doctor thought: “Mary was pregnant”.
Backward-shifted
b. Every doctor thought: “Mary is pregnant”.
Simultaneous
c. *Some doctors thought: “Mary was pregnant”, and
Some doctors thought: “Mary is pregnant”.
Mixed
Recall that the ULC rules out time concepts that involve future reference, such as
“tomorrow”. On the face of it, then, the time concept in (29) should be allowed.
(29)
At the time I am Mary’s doctor, be it now or before.
Suppose that Mary was seriously injured, and after being brought to the hospital, the
doctors at the E.R., Sue and Sally, were consulting each other. Sue was Mary’s doctor
at that time, and Sally was Mary’s doctor two years before that. For some reason, it
was relevant for Mary’s treatment whether she had been pregnant at any time in her
life. Sue thought: “Sure, she is pregnant now”, and Sally thought: “She was pregnant
when I was her doctor, two years ago”. Following the ULC-based approach, the
functional de re reading of (28) using the time concept in (29) should give rise to the
(iii)
John thinks Bill’s 40th birthday is in the past and that Mary was pregnant on that day. Bill’s
40th birthday is in fact the day of John’s thinking, which is today.
In this context, (1) is infelicitous. However, the ULC-based approach predicts it to be felicitous:
The indexical component of DAR requires inclusion of UT (in reality), and the time of Bill’s 40th
birthday does include the utterance time; the relative component provided by the ULC requires Bill’s
40th birthday not to be later than AT (in John’s mind), and it isn’t.
Heim (1994) therefore suggests that there is a prohibition on temporally neutral time concepts.
However, Bary & Altshuler (2015) show that once the relative component of DAR requires that the
time of Mary’s pregnancy include AT rather than include or precede it, this prohibition is redundant
and temporally neutral time concepts become harmless. Note that this is in line with (6).
15 I thank Danny Fox for pointing out the relevance of such examples. The same argumentation can be
done with quantification over times as in (iv) and with ellipsis as in (v).
(iv)
In the last few years, every time I met John he told me that he was sick.
(v)
Yesterday I met John and Mary. She told me she was sick, and he did too.
198
De re tenses and trace conversion
reading in (28c) in this scenario. However, no such reading is available.16
In other words, this situation is compatible with the predicted functional de re
reading for (28) given the time concept in (29), which can be described as follows:
(30)
For each doctor, the time of which she believed Mary to be pregnant is s.t.:
a. From the speaker’s perspective it precedes the “now”.
Indexical c.
b. From that doctor’s perspective it’s the time of her being Mary’s doctor
(which is not later than her “now”).
Relative c.
Given (6) no mixed reading is predicted: Since a de re past yields no backwardshifted reading, it yields no mixed simultaneous and backward-shifted one as well.
Can we maintain the ULC and rule out (28c)? A few ideas may come to mind:
i. Functional readings are not possible for tenses interpreted de re. That is,
the de re past cannot be bound by the universal quantification. This would
indeed solve the problem, but it is not clear why that would be the case: what
would allow such binding for non-de re tenses and prevent it for de re tenses?
ii. Temporally neutral time concepts are ruled out, as suggested by Heim
(1994: fn. 28). First, for that to help here we would have to say that the time
concept in (29) counts as temporally neutral even though it does encode at
least some temporal relation. Second, as Bary & Altshuler (2015) show,
ruling out temporally neutral time concepts is not necessary if the relative
component of DAR requires inclusion (see footnote 14).
These are indeed possible refutations of the argument presented here, but as I have
shown none of them is trivial. Combined with the stipulative nature of the ULC, the
data presented here might be a good-enough reason to rethink it.
5
Concluding remarks
To conclude, the current proposal has several advantages over previous de re analyses. First, the ULC assumed by Abusch (1997) can be dispensed with, as in Ogihara
16 It should be noted that the data here are problematic for pragmatic accounts of SOT phenomena such
as Gennari 2003. According to Gennari, simultaneous readings of past-under-past in English do not
require a different representation than backward-shifted readings, but result from a real embedded
past tense and a “superinterval property” of stative predicates which allows a past state to stretch up
to the present. Since the backward-shifted reading and the simultaneous reading are derived from
the same LF in this analysis, it is not clear why that LF would not admit a mixed reading of the two,
as in (28c). Even though Altshuler & Schwarzschild (2013) also assume only one representation
for English past-under-past sentences, their analysis differs from Gennari’s in important respects.
Because of that, the question of the vulnerability of Altshuler & Schwarzschild’s analysis to this
argument is more subtle and requires another discussion, which I will not pursue here.
199
Bar-Lev
1989. Second, simultaneous readings of past-under-past in Hebrew and the DAR
of present-under-past in English are both accounted for as instances of a de re interpretation of tenses, in contrast with Ogihara’s Copy-based approach. Third, the
unavailability of mixed functional readings of past-under-past sentences is naturally
accounted for, unlike in Abusch’s ULC-based approach.
As has been shown in the last section, these advantages result from the fact that
the current proposal predicts (6) whereas previous analyses do not.
One may ask what alternatives we have, if we are convinced that a de re analysis
of tenses is needed and that the Inclusion Generalization in (6) is correct. For example, my analysis relies heavily on the assumptions that tenses are quantificational and
that they can move: The interplay between the inner structure of the quantificational
tenses and the application of Trace Conversion to moved tenses enabled me to predict
(6). Is there a way to do without these assumptions?
For now, I can only think of one possible alternative, which assumes no tense
movement. Suppose that tenses are pronominal and have the following semantics (I
discuss here only PAST tenses for brevity, but this also extends naturally to PRES).
(31)
JPASTi, j Kg is defined only if g(i) < g( j). When defined, JPASTi, j Kg = g(i)
When the second index of tenses is locally bound, i.e., when they are evaluated with
respect to the local evaluation time, nothing special needs to be said.
(32)
λt1 John PAST2,1 think that λt3 Mary PAST4,3 be-pregnant
The problem begins where the second index of a tense is non-locally bound, namely
if instead of PAST4,3 in (32) we would have PAST4,1 . Let us adopt a concept-generator
mechanism as in Percus & Sauerland 2003 and Charlow & Sharvit 2014, and assume
that tenses that are non-locally bound are interpreted through the application of a
concept generator which is syntactically realized, as in (33).
(33)
λt1 John PAST2,1 think that λ G8 λt3 Mary [G8
PAST 4,1 ]
be-pregnant
G8 provides a suitable acquaintance relation between John and the time g(4). Now,
what if we assumed that acquaintance relations about times are so restricted that
any acquaintance relation which does not involve presentness is ruled out? In other
words, the only time with which we are sufficiently acquainted is our “now”. With
both the past and the future we are not sufficiently acquainted. So “today” and “this
month” are fine descriptions, but “yesterday” and “tomorrow” are not.
In a sense, this would be both an Upper Limit and a Lower Limit constraint.
But differently from Abusch’s (1997) ULC, it will not be done by constraining
the assignments of tense nodes, but rather by constraining acquaintance relations.
Therefore it will not affect cases of tenses that are locally bound in which no
200
De re tenses and trace conversion
acquaintance relations are involved, but only de re tenses. However, similarly to the
ULC, such a restriction on acquaintance relations seems unmotivated.
One possible motivation to prefer this kind of analysis over the one I suggested
in Section 3 is that the way PAST is defined in (14b) is problematic. This definition of
PAST renders it a non-conservative quantifier, i.e., (34a) and (34b) are not equivalent.
(34b) is in fact contradictory: t should be both in T and at the same time before some
time in T . Given that T is the set of times that include the evaluation time, there is
no time in T which precedes another time in T , so this yields a contradiction.
(34)
a.
b.
JPASTK(T )(P) = 1 iff ∃t[∃t 0 [T (t 0 ) ∧ t < t 0 ] ∧ P(t) = 1]
JPASTK(T )(T ∩ P) = 1 iff ∃t[∃t 0 [T (t 0 ) ∧ t < t 0 ] ∧ T (t) ∧ P(t) = 1]
This is problematic for local movements of PAST: In cases where two copies of PAST
appear in the scope of the same local evaluation time, the result would be a fatal
presupposition failure: The lower copy will presuppose inclusion of the evaluation
time whereas the higher copy will assert precedence with respect to it.
It has been suggested (Fox 2002: fn. 8, a.o.) that conservativity of natural
language determiners follows from the Copy theory of movement: Given the Copy
theory, non-conservative quantifiers would lead to trivial meanings. It has been noted
by Sportiche (2005), Hallman (2012) and Romoli (2015), that once we allow two
copies to be evaluated with respect to two different worlds, this connection breaks
down and non-conservative quantifiers may give rise to non-trivial meanings. This is
essentially what is assumed here: The two copies of PAST must be interpreted with
respect to different times, otherwise they would lead to trivial meanings.
Relatedly, it is assumed here that two copies can have two different binders: One
copy can be bound by the matrix local evaluation time and the other by the embedded
one. One may ask whether this is natural, and if we could find anything like that in
the nominal domain. I have no good answers to these questions. I thank Danny Fox
and Uli Sauerland for pointing that out, and leave it for further discussion.
Finally, the motivation of Ogihara (1989) for his Temporal Directionality Isomorphism, represented here in (23), is data from Japanese in which there are no
simultaneous readings for past-under-past. The solution Ogihara & Sharvit (2012)
suggest involves making the question of whether movement leaves a Copy behind
subject to crosslinguistic variation, a route which I tried to avoid.
To the extent that there are indeed no such readings in Japanese, in the view
adopted here it means that the structure where past has QRed is blocked. Note
that for Hebrew there is anyway still a need for an account of when the QRed
structure is available for past-uner-past sentences, namely the conditions under
which a simultaneous reading is available (see footnote 13); and for English too, the
unavailability of a present-under-past in-situ needs to be explained (see footnote 2).
201
Bar-Lev
References
Abusch, Dorit. 1997. Sequence of tense and temporal de re. Linguistics and
Philosophy 20(1). 1–50. doi:10.1023/A:1005331423820.
Altshuler, Daniel & Roger Schwarzschild. 2012. Moment of change, cessation
implicatures and simultaneous readings. In Emanuel Chemla, Vincent Homer
& Gregoire Winterstein (eds.), Sinn und Bedeutung (SuB) 17, 45–62. http:
//semanticsarchive.net/sub2012/.
Altshuler, Daniel & Roger Schwarzschild. 2013. Correlating cessation with double
access. In Maria Aloni, Michael Franke & Floris Roelofsen (eds.), 19th Amsterdam Colloquium, 43–50. http://www.illc.uva.nl/AC/AC2013/Proceedings/.
Bary, Corien & Daniel Altshuler. 2015. Double access. In Eva Csipak & Hedde
Zeijlstra (eds.), Sinn und Bedeutung (SuB) 19, 89–106.
Charlow, Simon & Yael Sharvit. 2014. Bound ‘de re’ pronouns and the LFs of
attitude reports. Semantics and Pragmatics 7(3). 1–43. doi:10.3765/sp.7.3.
Comrie, Bernard. 1986. Tense in indirect speech. Folia linguistica 20(3–4). 265–296.
doi:10.1515/flin.1986.20.3-4.265.
von Fintel, Kai & Irene Heim. 2011. Intensional semantics. Manuscript, Massachusetts Institute of Technology, Spring 2011.
Fox, Danny. 2002. Antecedent-contained deletion and the copy theory of movement.
Linguistic Inquiry 33(1). 63–96. doi:10.1162/002438902317382189.
Gennari, Silvia P. 2003. Tense meanings and temporal interpretation. Journal of
Semantics 20(1). 35–71. doi:10.1093/jos/20.1.35.
Grønn, Atle & Arnim von Stechow. 2010. Complement tense in contrast: The SOT
parameter in Russian and English. Oslo Studies in Language 2(1). 109–153.
Hallman, Peter. 2012. Prospects for a syntactic analysis of conservativity. UCLA
Working Papers in Linguistics 17. 103–108.
Heim, Irene. 1994. Comments on Abusch’s theory of tense. In Hans Kamp (ed.),
Ellipsis, Tense and Questions, Dyana-2, Deliverable R2.2.B.
Jespersen, Otto. 1924. The philosophy of grammar. London: Allen and Unwin.
Klecha, Peter. 2014. Modality and embedded temporal operators. Manuscript, Ohio
State University.
Kratzer, Angelika. 1998. More structural analogies between pronouns and tenses.
In Devon Strolovitch & Aaron Lawson (eds.), Semantics and Linguistic Theory
(SALT) 8, 92–110. Ithaca, NY: Cornell University. doi:10.3765/salt.v8i0.2808.
Kusumoto, Kiyomi. 2005. On the quantification over times in natural language.
Natural Language Semantics 13(4). 317–357. doi:10.1007/s11050-005-4537-6.
Ogihara, Toshiyuki. 1989. Temporal reference in English and Japanese. University
of Texas at Austin PhD dissertation.
Ogihara, Toshiyuki & Yael Sharvit. 2012. Embedded tenses. In Robert I. Bin-
202
De re tenses and trace conversion
nick (ed.), Handbook of Tense and Aspect, 638–668. Oxford University Press.
doi:10.1093/oxfordhb/9780195381979.013.0022.
Partee, Barbara Hall. 1973. Some structural analogies between tenses and pronouns
in English. The Journal of Philosophy 70(18). 601–609. doi:10.2307/2025024.
Percus, Orin. 2000. Constraints on some other variables in syntax. Natural Language
Semantics 8(3). 173–229. doi:10.1023/A:1011298526791.
Percus, Orin & Uli Sauerland. 2003. On the LFs of attitude reports. In Matthias
Weisgerber (ed.), Sinn und Bedeutung (SuB) 7, 228–242. http://semanticsarchive.
net/Archive/TE3NGVlY/.
Romoli, Jacopo. 2015. A structural account of conservativity. Semantics-Syntax
Interface 2(1). 28–57.
Schlenker, Philippe. 1999. Propositional attitudes and indexicality: A cross categorial approach. Massachusetts Institute of Technology, Department of Linguistics
and Philosophy, PhD dissertation.
Schlenker, Philippe. 2004. Sequence phenomena and double access readings generalized: Two remarks on tense, person, and mood. In Jacqueline Guéron &
Alexander Lecarme (eds.), The Syntax of Time, Cambridge, MA: MIT press.
Sportiche, Dominique. 2005. Division of labor between merge and move: Strict
locality of selection and apparent reconstruction paradoxes. Manuscript, UCLA.
von Stechow, Arnim. 1995. On the proper treatment of tense. In Mandy Simons &
Teresa Galloway (eds.), Semantics and Linguistic Theory (SALT) 5, 362–386.
Ithaca, NY: Cornell University.
Stowell, Tim. 1995. What do the present and past tenses mean? In Pier Marco
Bertinetto, Valentina Bianchi, James Higginbotham & Mario Squartini (eds.),
Temporal Reference: Aspect and Actionality 1, 381–396. Torino: Rosenberg &
Sellier.
Stowell, Tim. 2007. The syntactic expression of tense. Lingua 117(2). 437–463.
doi:10.1016/j.lingua.2005.08.003.
Moshe E. Bar-Lev
Language, Logic and Cognition Center
The Hebrew University of Jerusalem
Mount Scopus
91905 Jerusalem
Israel
[email protected]
203